Instrumental variable methods Sample Clauses

Instrumental variable methods. When examining post randomisation variables, such as dose, confounding may occur even in the randomised clinical trial scenario. The instrumental variable design offers, under certain assumptions, the opportunity to tame confounding bias, without directly observing all confounders. The instrumental variable approach is well developed in the context of linear regression (Xxxxxxxxxx 2010) and generalised linear models (Xxxxxxxxxxxx and Goetghebeur 2003). Tchetgen Tchetgen (Tchetgen Xxxxxxxx, Xxxxxx et al. 2015) develop the instrumental variable approach in a survival context. He describes two strategies utilising Aalen’s additive hazard model (Aalen 1989). The first approach is analogous to that commonly used for estimation in linear models; the two stage least squares approach. Here, the fitted value from the first stage regression of exposure on the instrumental variable is entered in place of the exposure in the second stage hazard model to give an estimate of causal effect. The second approach is the two stage residual inclusion (TSRI); here the residual from the first stage regression of the exposure on the instrumental variable is added as a confounding control covariate to the additive hazards regression model for the exposure effect. This work utilises the second approach. A correct instrumental variable analysis rests on three fundamental assumptions (Xxxxxxx and Savitz 1997, Greenland 2000, Xxxxxxx, Xxxxxxx et al. 2006)